The dominant approach to studying historical race-related fertility differences has been to limit samples to first-married and younger women. We argue that studying historical race-related fertility differences in the context of remarriage is also important: remarriage and fertility patterns are both rooted in the biosocial conditions that produce racial disparities in health. We employ a multiple causes framework that attributes variation in fertility patterns to voluntary limitation and involuntary factors (infecundity/subfecundity). We use data from the 1910 Integrated Public Use Microdata Series and estimate zero-inflated negative binomial models that simultaneously distinguish those who are infecund (vs. fecund) and estimate the number of remarital births among the fecund. Our approach allows us to evaluate historical remarital (in)fertility differences, accounting for marital, socioeconomic, and geographic influences on fecundity and fertility, while empirically accounting for the influence of children “missing” from the household due to mortality and fostering/aging out. Consistent with past studies that emphasized poorer African American health as a major influence on involuntary infertility, we find that African American women were more likely than white women to be in the always-zero (infecund) group and to have fewer remarital births. Supplemental analyses nuance these findings but indicate that these results are robust. Overall, we find support for a multiple-causes perspective: while the findings are consistent with the adoption of deliberate fertility control among urban and higher-status women at higher parities, remarital fertility differences in 1910 also reflected greater infecundity/subfecundity among subgroups of women, especially African American women.
U.S. marital fertility patterns at the turn of the twentieth century differed considerably by race.1 Fertility rates among white married women in 1900 were in decline, largely due to voluntary, parity-specific control (contraception and abortion) (Guest 1981; Hacker 2003; Tolnay et al. 1982). In contrast, fertility rates among African American married women in 1900 were in decline due to involuntary infecundity and subfecundity2 as well as voluntary control (Cutright and Shorter 1979; Farley 1970a; Masnick and McFalls 1976; McFalls and McFalls 1984; Tolnay 1981, 1983, 1987, 1989).3 In the early twentieth century, African American women had higher fertility rates but also higher rates of childlessness than white women (Farley 1970a; McFalls and McFalls 1984), likely because African American women (more than white women) were subject to poor living standards, diseases, and childbirth-related morbidity, with severe consequences for fecundity, parturition, and infant and child survival (Costa 2004; Cutright and Shorter 1979; Elman et al. 2015; Farley 1970a, b; Howard 1921; McFalls and McFalls 1984; Wertz and Wertz 1977). However, race differences in involuntary infecundity/subfecundity remain poorly understood in the U.S. historical fertility literature.
A second, less-understood race-related historical demographic difference involves remarriage patterns around 1910 (London and Elman 2001; Preston et al. 1992).4 More often widowed and divorced, African American women were twice as likely as same-aged white women to remarry (Glick 1949; Kramarow 1995). This difference is important because fertility patterns differ in first marriages versus remarital unions (Cohen and Sweet 1974; Thornton 1978), often for health reasons associated with race (Griffith et al. 1984). Most historical studies of race-related fertility differences, however, have studied only samples of women deemed likely to be in first marriages (see Tolnay 1984).5 Thus, sample selectivity in past studies may have inadvertently obscured understanding of historical race-related differences in fertility. Specifically, the experience of remarried women in the early twentieth century was distinctive because the joint likelihood of infecundity/subfecundity and remarriage increased with age among all women, and especially African American women. Put another way, widowhood/remarriage and childbearing overlapped in historical women’s lives and were especially likely to co-occur among African American women.
We use 1910 Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2010) data to examine remarital births in a sample of women aged 15–49 at remarriage. Our study is important because to date, remarriage has not been adequately addressed in historical demographic research on race and fertility. We use multivariate zero-inflated negative binomial (ZINB) modeling to simultaneously estimate the log odds that remarried women were infecund (vs. fecund) and the number of remarital births among fecund women (Guinnane et al. 2002). ZINB models allow us to examine race-related differences in fecundity and fertility, controlling for a broad range of marital, socioeconomic, household, and geographic variables. Importantly, we empirically account for the presence of children from women’s and men’s prior unions as well as for children “missing” from households due to mortality and fostering/aging out. Net of these other influences, we expect that African American women will (1) be more likely to be in the infecund group, and (2) have fewer remarital births (i.e., exhibit subfecundity). Such findings will provide novel evidence of race-related fertility differences in 1910.
Fertility: A Multiple-Causes Framework
Demographic studies of the United States in the mid–twentieth century found—as predicted by demographic transition theory and its variants—that early twentieth century African American and white women who were literate, were urban, and/or lived in communities marked by an increasing prevalence of industrial versus agricultural jobs had lower fertility (Engerman 1977; Farley 1970b). However, these same studies identified anomalous patterns. African American women had higher fertility rates than white women in 1900, but their falling rates closed the black-white fertility gap by the 1930s (Farley 1970a, b; Masnick and McFalls 1976). Moreover, accumulating research had found that the major leg of the African American fertility decline that commenced after 1900 occurred under circumstances opposite to those articulated in demographic theories. Engerman (1977:118) wrote:
The sharp decline in fertility among U.S. blacks was apparently not preceded by a large-scale decline in mortality as was the case with other populations. . . . it occurred while the black population was still primarily rural and agricultural and was located in a region of the country not undergoing significant industrialization or large-scale urbanization.6
By the 1980s, context-specific historical studies employed sophisticated techniques, such as Coale and Trussell’s (1974, 1978) methodology, which allows indirect estimation of the extent of deliberate marital fertility control in a population. Puzzling findings continued to emerge. Native white marital fertility in 1900 was lowest in urban areas, especially in the industrializing North (Guest 1981; Tolnay et al. 1982; more recently, see Hacker 2003; Morgan 1991). African American women’s marital fertility was also low in the North and was exceedingly low—perhaps even below replacement level (Engerman 1977:124)—in urban locations across U.S. regions (Farley 1970a, b; Tolnay 1981). Consistent with these findings, evidence suggested that urban and rural white women and urban African American women adopted voluntary fertility control methods well before 1900 (Guest 1981; Tolnay 1983, 1987; Tolnay et al. 1982). But surprisingly, estimates derived from application of the Coale-Trussell method suggested that southern rural African American and white women in 1900 exhibited only minimal voluntary, parity-specific control; they resembled natural fertility populations (Tolnay 1983, 1987). Their high fertility likely reflected the fact that the southern rural population in the early twentieth century still largely toiled in farming, often in tenant farm households performing labor-intensive agriculture that required large family pools of workers (Whatley 1987; Wright 1986).7
Although studies using the Coale-Trussell method as a group reported U.S. fertility patterns that were strikingly similar by race across regions, which is consistent with the notion of context-specific, voluntary limitation in demographic theories, findings also challenged extant theories. White rural married women, including those living in the South, were somewhat further along in adopting voluntary control by 1900; African American rural married women did not widely adopt such practices until after this time (Tolnay 1981, 1987).8 This suggested that African American fertility decline in the early twentieth century had commenced in rural portions of the nation and in the South, where fertility stood at near-natural levels, voluntary control was limited, and economic incentives for large families remained high (Farley 1970a; Tolnay 1981, 1983, 1984, 1987). How could this happen?
To address this anomaly, a second school of thought proposed that involuntary factors—primarily infertility due to poorer health and lower living standards among African Americans—had significantly contributed to observed patterns (Farley 1966, 1970a; Tolnay 1987). Poor maternal health can go hand in hand with and sustain a high prevalence of infecundity, miscarriage, childlessness, and subfecundity in populations with otherwise high fertility (McFalls and McFalls 1984). The reverse—high fertility leading to poorer maternal health and infecundity—is possible as well.
The demographers of the 1970s and 1980s who raised the issue of historical race differences in involuntary influences on fertility based their arguments on a long arc of medical and public health research. Existing studies documented complex biosocial effects on fertility (Pearl 1936) and higher rates of African American than white childlessness, maternal morbidity, infertility, and infant and child mortality (Brandt 1985; Farley 1970a, b; Howard 1921; Wertz and Wertz 1977; more recently, see Costa 2004). Studies of the mid-twentieth century found a substantial prevalence of subfecundity among even relatively healthy U.S. whites (e.g., Freedman et al. 1959). Decades later, many demographers argued that poorer living standards, higher disease prevalence, and limited access to safe medical and obstetric care, especially among historical rural African American women, led to poorer fertility outcomes (Ewbank 1987a, b; Suliman 1983; Tolnay 1983).
Inadequate housing, poor sanitary conditions, limited food supplies, and/or harsh work conditions in the early twentieth century contextualized women’s considerable risk of involuntary infecundity/subfecundity. Federally supported household food consumption studies from the 1890s onward reported that the typical African American diet among northern urban and southern rural farm families was poor enough to produce chronic malnutrition (Dirks 2003; Dirks and Duran 2001). Moreover, African Americans were far more likely to be farm tenants (vs. autonomous farm owners) than were whites (Alston and Kaufmann 1998; Elman et al. 2015). For farm tenants, chronic malnutrition was further punctuated by seasonal bouts of near starvation; tenancy contracts settled annually in the fall, and meager year-end surpluses or debt left few households the resources needed to purchase food over the winter. African American women were especially susceptible to poor nutrition: vitamin D deficiency and rickets in childhood and adolescence affected pelvis size, and hence later childbearing risk (Carson 2008; Cutright and Shorter 1979; Fuller 2000; McFalls and McFalls 1984). Poor living standards also increased women’s susceptibility to prevailing infectious diseases, such as tuberculosis and malaria (Humphreys 2001; McFalls and McFalls 1984).
With regard to specific diseases, tuberculosis mortality among African Americans rose dramatically after 1880; by 1920, their rates were two to four times higher than rates among whites (Public Health Reports 1923; see also Beardsley 1990; McFalls and McFalls 1984). Louis I. Dublin (n.d.), statistician for the Metropolitan Life Insurance Company, analyzed African American policyholder data for 1920, which represented 20 % of the black population, and noted disproportionately high tuberculosis death rates among 10- to 14-year-old African American girls. McFalls and McFalls (1984) suggested that historical tuberculosis rate differences significantly shaped race-related infertility differences in this era. They noted that tuberculosis, when contracted at young ages, is more often marked by extrapulmonary lesions, that such lesions are less often dormant, and that survivors of childhood and adolescent tuberculosis are at higher risk of infertility and subfecundity. Malaria was also endemic in many portions of the United States in 1900 and across the U.S. South until the 1940s (Humphreys 2001). Even with acquired immunity to local strains, women could suffer severe anemia, childbirth complications, and fetal loss if reexposed during pregnancy (Desai et al. 2007; Lucas 2013; Rose 1989). According to the earliest data available (Maxcy 1923), African American malaria mortality rates were at least twice as high as white rates during the period 1919–1921. Venereal disease and hookworm also affected African American and white fertility rates; differences in TB (McFalls and McFalls 1984), hookworm (Bleakley and Lange 2009) and venereal disease (Tolnay 1989; Tolnay and Glynn 1994) prevalence may have accounted for small portions of the race-related differential fertility decline by 1940.
Exacerbating the direct effects of higher African American rates of tuberculosis, malaria, hookworm, typhoid fever, venereal disease, pneumonia, and numerous combinations of diseases on childbearing was limited African American access (if any) to medicines (such as quinine) or to effective medical and obstetric care (Beardsley 1990; Brandt 1985; Costa 2004; Humphreys 2001; Wertz and Wertz 1977). The greater racial barriers to treatment and prevention in this era amplified individual disease risk and severity (Thomas 2011).
Evidence has since mounted that involuntary influences on fertility played a significant role in African American marital fertility patterns in the early twentieth century. For example, Cutright and Shorter (1979; also see Farley 1966) uncovered voluntary and involuntary influences in parity progression ratios among U.S. women born between 1867 and 1935. Earlier birth cohorts of nonwhite women appear to have faced a fecundity threshold: they were more often childless but progressed to a higher parity with each birth if a second birth occurred. In contrast, in earlier birth cohorts, more white than nonwhite women stopped at first and second births, with lower odds of childbearing with each subsequent birth. This latter pattern is more indicative of voluntary control.
Tolnay (1989) and Tolnay and Glynn (1994) advanced a multiple-causes framework for their historical studies of race and fertility. This framework suggested that variation in fertility in the early twentieth century was a function of voluntary limitation, associated with regional socioeconomic development, and involuntary infecundity/subfecundity attributed to health impairment. Tolnay and Glynn (1994) found strong evidence of multiple causes in the fertility patterns of African American and white women as late as 1940; the county-level child-woman ratios of African American and white women in the southern states reflected both voluntary practices (indicated by development variables) and involuntary factors (indicated by infant mortality and local syphilis rates).
More recently, Elman et al. (2015) used 1910 IPUMS data to examine race-related differences in the transition to a first birth among first-married women in marriages of three years or less. This marital context maximizes the physiological, behavioral, and social conditions for at least one birth to occur and minimizes limitation behaviors. Stated otherwise, limitation in this era would motivate stopping at higher parities more than delaying a first birth (Morgan 1991). They found, net of county industrial and agricultural development, as well as a range of other demographic and contextual controls, that first-married African American women were significantly less likely than white women to have at least one birth over the first three years of marriage. Consistent with Cutright and Shorter (1979), African American women appeared to face a fecundity threshold associated with involuntary factors.
Studies of race-related fertility differences in the early twentieth century have clearly demonstrated that both voluntary and involuntary factors were influential. Thus, we adopt the multiple-causes framework for our analysis. However, unlike previous multiple-causes studies, which examined only first- or ever-married women, we focus on remarried women.
Marital Disruption, Remarriage, and Fertility
Race-related differences in marital (as well as fertility) patterns existed in the historical population that we study. Largely due to mortality, only 38 % of African American first marriages in the early twentieth century retained both partners to the end of women’s reproductive years (Preston et al. 1992).9 In cross-section, in 1910, 17 % of all African American women had remarried, compared with 7 % of all white women, with rates rising in each age group (Glick 1949).10 Most young U.S. adults in 1910 who had experienced marital disruption had remarried (Glick 1980), although more household resources or personal wealth could accelerate or delay the timing of remarriage (van Poppel 1995). In this era, first marriages and remarriages were motivated as much or more by family economy needs and circumstances than individual (personal) desires (Cohen and Sweet 1974; Elman and London 2002; Grigg 1977; van Poppel 1995). A spousal loss imposed severe economic penalties on the family economy. Unless the widowed or divorced had older children who could contribute to household maintenance and/or economic support, remarriage was likely (Chudacoff and Hareven 1979; Hareven and Uhlenberg 1995; Riley 1991). Additionally, health and perhaps fecundity were factors in (re)marital partner choices.
(Re)marriages also reflected contextual factors, such as the size and composition of local marriage markets. Rural African Americans, especially in the South, were exposed to marriage markets in which potential partners had strong economic incentives to (re)marry because the land tenancy system penalized lone or unmarried workers (Bloome and Muller 2015; Brannen 1924; Woodman 1995). Indeed, African American and white (re)marriage patterns in 1910 responded to rural development in different ways. Shrinking average farm sizes and rising land cost in the Midwest and South were associated with rising farm tenancy versus ownership (Fite 1984; Gratton 1987; Wright 1986). This encouraged white males to delay family formation to obtain tools, animals, and/or land needed to enter higher-status cash tenancy or ownership arrangements. However, these same pressures encouraged African Americans and poor whites to become croppers or share tenants, and to start families earlier in life (Bloome and Muller 2015; Landale 1989; Tolnay 1984). Yet, African Americans were less likely to ever move up the agricultural ladder; their higher remarriage rates not only reflected younger ages at first marriage and greater marital disruption due to mortality and divorce but also higher odds of lifetime farm tenancy versus ownership (Alston and Kauffman 1998; Elman et al. 2015; Fite 1984; Woodman 1995).
Variation in the cultural acceptability of remarriage was also important. Higher mortality among African Americans meant that more of their potential remarital partners had been previously married. In addition, African American men in 1910 were significantly more likely than white men to marry previously married women, even when women had children from a previous union (London and Elman 2001). This pattern of greater African American stepfamily formation via remarriage, in addition to higher overall remarriage rates, likely reflected a greater cultural acceptability of remarriage following widow(er)hood and divorce in the African American community of the time (Clarkwest 2006; Glick 1949; London and Elman 2001; Pagnini and Morgan 1996).
Both marital disruption and remarriage can shape fertility (Glick 1980), although the direction of the influence might vary (Wineberg 1990). Marital dissolution followed by remarriage can raise the odds of remarital childlessness (Davis 1982; DeLong and Sells 1977). Perhaps, net of the effect of delayed or interrupted exposure to the risk of childbearing, women are motivated to limit remarital fertility because marital disruption accustoms them to having fewer children and/or pursuing work or other activities. Women also might limit remarital fertility to devote time and personal resources to new spouses and/or stepchildren. However, remarriage often leads to renewed childbearing as couples cement the new marriage with children or correct for the subfecundity of their first spouse (Thornton 1978). With regard to the possibility of cementing remarriage with children, the number of children born in first marriages does not appear to affect remarital fertility; women do not count or “trade off” children from a first marriage against total desired fertility. Rather, they have children to “validate the marriage” (Griffith et al. 1985:79).
These studies focus on mid–twentieth century childbearing—generally, post-divorce—and thus may not be applicable to the historical period we study. However, we believe that childbearing in a previous union should have affected early twentieth century remarried women even less than mid-century remarried women (i.e., it should not have reduced remarital fertility). Childbearing was an obligation of marriage in the earlier portion of the twentieth century; social and institutional forces upheld more traditional gender and marital norms (Davis 1982). Additionally, still-high early twentieth century mortality rates, especially among African Americans, meant that previously widowed women likely had firsthand experience with the death of a child as well as their spouse, and thus had incentive to renew childbearing upon remarriage. A final consideration that may have supported childbearing among remarried women is the marital status of their husband. Among remarried women who married previously unmarried men, the desire for—or expectation of—having children likely would have been particularly strong.
Evidence also suggests that remarriage may be strategic if a previous spouse was infertile; remarried women’s prior spouses may not have been able to father children because of old age or because of long illnesses prior to death or divorce (Cohen and Sweet 1974). However, infertility in first marriages might instead have been due to women’s own poor health. Approximately 30 % of mid-century women who were childless in first marriages remained childless in a subsequent marriage (Griffith et al. 1984). Consistent with the multiple-causes framework, this statistic partly reflected voluntary factors (indicated by higher education, employment, and urban location) and involuntary factors (such as undiagnosed infecundity or subfecundity) (Griffith et al. 1984).
We examine race-related differences in remarital fertility in 1910, adopting the multiple causes framework to take both voluntary and involuntary influences on fertility into account. Mid–twentieth century studies have found that upon remarriage, there was a strong desire to cement marriages to please new spouses or in-laws and/or start new family lines. This motivation was likely as strong or stronger in 1910. African American women’s higher remarriage rates—at younger mean ages at remarriage—should also have increased exposure to the risk of remarital childbearing. There is no reason to believe that African American women in this era were less likely than white women to desire to cement marriages. Additionally, the desire for a child should have been as great or greater among African American than white women because of their disproportionate concentration in agriculture and farm tenancy, where children were economic contributors, and because of the higher cultural acceptability of remarriage among African Americans.
It is also important to consider that African American women’s fertility rates were high in 1910—other than in the North and urban areas (we adjust for these factors)—with rural rates resembling rates found among natural fertility populations (Tolnay 1983). If no race difference in fecundability is present, ceteris paribus, we would expect to find a higher likelihood of a first remarital birth and equal if not higher overall remarital fertility among African American women relative to white women. However, multiple-causes research provides evidence that race-related differences affecting fecundity and fertility were present. Thus, we hypothesize that African American women were more likely than white women to be infecund (vs. fecund) net of other factors. Similarly, net of other factors, we hypothesize a lower number of remarital births among fecund African American than white women given that infecundity and subfecundity can emerge over the life course in relation to enduring poverty, illness, and childbirth complications. We acknowledge one caveat with respect to this second hypothesis. Consistent with the multiple-causes framework, in some circumstances, fecund African American women with proven remarital fertility might be expected to have a greater number of higher-order births than similar white women, who potentially faced fewer pressures to continue childbearing and had more access to voluntary methods of fertility control.
Data and Methods
We use data from the 1910 IPUMS, a 1-in-70 sample of households taken from microfilm records of the 1910 U.S. Census of Population (Ruggles et al. 2010). The 1910 census is the only historical census that includes information about remarriage, current marital duration, and women’s fertility (i.e., children ever born and surviving).11 We link household child and parent records; we identify remarital births by allocating biological (nonsocial) children to a parent’s prior or current union by comparing each household child’s age with the duration of their parent’s current union (read more about this in the upcoming section, Focal Variables). We further account for mothers’ children who are missing from the household due to death and/or fostering/aging out (i.e., home-leaving). Our approach allows us to study remarital fertility differences with individual-level, cross-sectional data in order to investigate hypotheses related to potential race-related differences in remarital fecundity and fertility.
In 1910, census enumerators asked married persons whether their current union was a first or higher-order marriage. We sample remarried women of any age whose age at remarriage was 15–49 and use an IPUMS-constructed indicator—SPLOC—that identifies a spouse in the household. We link to each remarried woman’s record her nonsocial children's and spouse's information, including his marriage number and indicators of his children from prior marriages who reside in the household. We drop a small number of remarried women with missing data on their own or their spouse’s marriage number. Thus, the analytic sample includes all white and African American (i.e., black and mulatto) women who were aged 15–49 at the time that their current remarriage was contracted, regardless of whether the woman was remarried to a first-married or remarried man. We exclude all women who were not enumerated as white, black, or mulatto, and set the upper age at remarriage at 49 because this age range includes most potentially fecund women who were at risk of a birth within their current remarital union.
With the exception of age at remarriage, we employ listwise deletion of cases with missing data. For age at remarriage, we calculate age-specific mean ages at remarriage for those with complete data and use those values to impute age at remarriage for those of a given age who were missing age at remarriage. Overall, we delete 2.19 % (unweighted; N = 391) because of missing data. The final analytic sample with no missing data includes 17,442 remarried women.
IPUMS-constructed variables for parents identify the record number of each child in the household and whether they were biological versus social children. For each mother’s record, we create variables to indicate each biological child’s age; subtracting each child’s age from the census-provided measure of length of current marriage allows for the identification of the number of surviving coresident children born during the current remarital union. We use a continuous measure of the Number of Remarital Births, ranging from 0 to 12, as the dependent variable in our analyses.
Race is our primary independent variable: African American (i.e., black or mulatto) (yes = 1) versus white.
Age at Remarriage is derived by subtracting the woman’s marital duration from her current age. This continuous variable ranges from 15 to 49 years because of the constraint we impose for inclusion in the analytic sample. Duration of Current Marriage is a continuous variable provided by the census. Remarital Configuration is identified by a dichotomous variable that measures remarried woman plus husband in first marriage (= 1) versus both remarried. As noted earlier, for each woman, we identify biological children and their ages, attach this information to the woman’s record, and compare children’s ages with current remarriage duration to determine whether each child was born prior to or after the current union was contracted. Children older than the duration of a woman’s current remarriage are considered children from a prior union. Number of Children From Own Prior Union is a count of the number of the woman’s own children from a prior union and is categorized as 0, 1–2, or 3+. Similarly, for spouses, we subtract the age of each own household (nonsocial) child, if any, from the measure of length of current marriage to derive a dichotomous indicator for Spouse Has Child From Prior Union in the household (yes = 1).
Literacy, occupational prestige, and tenancy variables indirectly index women’s differential access to nutrition, housing, and medical care resources. The 1910 IPUMS does not provide information about an individual’s level of educational attainment or income. However, Literacy represents a reservoir of human and social capital attainment, which in this era occurred early in the life course, prior to adult family formation. Highest Kin Prestige Score is an indicator of family socioeconomic status. The 1910 IPUMS reports an occupational prestige score for each employed person, measured by Duncan’s (1961) socioeconomic index, based on 1950 occupational codes and prestige rankings. Occupational prestige is highly correlated with job skill and access to opportunity structures of support, such as social networks and community and public resources. Because multiple household members tended to work in this era, with the occupational prestige of each member rising and falling over the life course, the highest score found among all working members of a coresidential kin group is assigned to each family member in the household. This method is similar to procedures of other studies that assume that family members derive status from the head of household or main breadwinner. Prestige scores range from 0 to 96, with higher scores signifying higher prestige. We combine two household-level variables to indicate farm tenancy: the first denotes farm residence, and the second indicates nonownership of dwelling. Tenancy is indicated by a dichotomous indicator of residence on an unowned farm (yes = 1).
Based on an IPUMS-derived variable, dichotomous variables indicate City (yes = 1) and Town (yes = 1) versus rural (population < 2,500) residence. Indicators for geopolitical region include North (yes = 1), Midwest (yes = 1), and West (yes = 1) versus South.
Controls for Missing Children
The 1910 census recorded, for ever-married women, the number of children ever born and surviving, respectively. For each woman, we derive the number of Children Missing Due to Death by subtracting her reported number of surviving children from her reported number of children ever born. The number of Children Missing Due to Fostering/Aging Out is calculated as the difference between each woman’s reported number of surviving children and the number of her biological children who were enumerated in the household.
We present descriptive and multivariate regression models. Multivariate models estimate race-related differences in fertility and include the marital, socioeconomic, geographic, and missing child controls described earlier. Our dependent variable—the number of remarital children born—is a count variable. The Poisson parametric distribution is most commonly used to model a count and assumes equality of the conditional mean and variance. However, some counts are overdispersed—that is, in this case, they include many or “excess” zeros (i.e., childless women)—leading to violation of this assumption (Guinnane et al. 2002:16). One way to address excess zero births is to consider whether the count is drawn from two different regimes of fertility. Thus, we test and report results from a splitting regression model that treats the Number of Remarital Births as the outcome of two processes or regimes: a logistic process that identifies an “always zero” (i.e., infecund) group, and a count process that models the number of births—including zero births—among presumably fecund women (Guinnane et al. 2002). To test the fit of the regression models, we compared zero-inflated with simpler count-only models. In all cases, a Vuong test indicated that the splitting regression model best fit the data. Additionally, because a significant α parameter indicated that the Poisson model exhibited overdispersion, we report results from the ZINB model. All descriptive analyses are weighted using the IPUMS-provided variable PERWT. Multivariate analyses are unweighted. We describe in upcoming text the supplemental analyses we conducted.
The aforementioned procedures allow us to (1) identify children born within remarital unions; (2) estimate the effect of race on infecundity and variation in the number of remarital births among fecund women; and (3) control for a broad range of contextual influences on fertility, as well as for the number of children who are missing from households. Doing the latter is important because African American parents may have “lost” more children from households than white parents because of higher rates of child mortality (Preston and Haines 1991) and fostering (McDaniel 1994), as well as earlier ages of leaving home (Ruggles 1994). This analytic approach is more appropriate for studying differences in remarital fertility than overall levels of remarital fertility because it is not possible to know whether missing children were from first or higher-order marriages. Although we cannot know whether a dead child is from a prior union or the current remarital union—nor exactly why a child is not present in the household—we can count the discrepancies and empirically account for them in our models in order to rule them out as explanations for race differences.
Table 1 describes the analytic sample. Approximately 21 % of remarried women were African American, which is an overrepresentation and indicative of the disproportionately high rates of remarriage among African Americans in this historical period. Women’s average age at remarriage was 31.23 years, which indicates that they had, on average, considerable childbearing potential at the time of remarriage. The average duration of their current remarriage was 11.60 years (range = 0–58 years), which suggests that they had, on average, considerable exposure to the risk of remarital childbearing.
As shown in Table 1, there is a significant race difference for each of the variables except number of children missing due to fostering/aging out. These race differences signal potentially countervailing influences on fecundity and remarital fertility. For example, white women were significantly older than African American women at remarriage, which might have reduced childbearing risk; however, they were remarried for significantly longer durations, which might have increased risk. Additionally, white women, relative to African American women, were more likely to be literate, to live in a high-occupational-prestige household, and to live in a town or city. These factors might have fostered a desire or provided the means to voluntarily limit fertility. However, white women were also more likely to have a husband in a first marriage and less likely to have a spouse’s child from a prior union in the household, which might have provided motivation to cement the current remarriage.
Beyond these race-related differences, African American women experienced more child deaths, which may have influenced their motivation to have additional children. Other factors also tended to favor higher remarital fertility among African Americans, such as their substantially higher residence in rural areas, in the South, and on tenant farms. These contextual circumstances are generally predictive of higher fertility.
Remarital Births by Race
Table 2 presents estimates of remarital births by race. African American women were substantially more likely than white women to have had zero remarital births (68.07 % vs. 58.41 %). Compared with white women, the percentage of African American women with 1–5 remarital births was about equal or lower; African American women were slightly more likely than white women to have 6 and 7–12 births, respectively. The observed distributions differ significantly by race. Taken together, the results suggest a racial crossover: relative to white remarried women, African American remarried women were less likely to have had a first remarital birth as well as lower-order births (1–5), but they were more likely to have had higher-order (6+) remarital births. These results support a multiple-causes interpretation; they suggest that an involuntary infecundity/subfecundity threshold among African Americans in the context of greater voluntary control among whites at higher parities might partially account for race differences in remarital fertility in 1910. We examine these patterns further in a series of multivariate regression analyses that account for the influences of marital, socioeconomic, geographical, and missing child variables to the extent possible with historical census data.
Remarital Births by Race Accounting for Missing Children
Table 3 presents a multivariate ZINB regression analysis of the number of remarital births among remarried women aged 15–49 at remarriage. Model 1 in Table 3 does not include the two control variables we developed to account for children missing from the household due to death and fostering/aging out; Model 2 includes these two control variables. As shown in Table 3, the addition of the controls for missing children does not substantially alter the observed associations; those associations that were significant in Model 1 remained significant in Model 2. We emphasize interpretation of the results from Model 2.
Focusing first on the African American coefficients, as shown in Model 2, net of a broad range of influences on fecundability and fertility, the log odds of being in the infecund group (i.e., the zero inflation portion of the model) were significantly higher among African American than white women. This finding is consistent with childlessness findings in Table 2. Among fecund remarried women (i.e., the count portion of the model), African American women had significantly fewer remarital births than white women.
Marital factors also shaped infecundity and the number of remarital births among fecund women in expected ways. Reduced exposure to the risk of childbearing due to older age at remarriage significantly increased the log odds of being in the infecund group and decreased the number of remarital births among fecund women. The opposite pattern is observed for duration of remarriage. Longer durations of remarriage significantly decreased the log odds of being in the infecund group, although the significant, positive coefficient on duration squared indicates a reduction in the effect at longer durations. Longer remarital durations increased the number of remarital births among fecund women but did so to a lesser extent over time, as indicated by the significant, negative coefficient on duration squared. The lower log odds of being in the infecund group at longer remarital durations may indicate that most women, even some subfecund women, eventually bore a child with long-term exposure to sexual intercourse. The nonlinearity in the effect of marriage duration on the number of remarital births among fecund women may partly capture a biological effect of aging on fecundability. Taken together, these results suggest that a later age at the dissolution of the women’s prior marriage and/or a greater length of time between marriages imposed a time penalty on remarital fertility or otherwise predisposed women toward having fewer children.
Beyond exposure to the risk of childbearing, remarital configuration mattered. Compared with couples in which both spouses had remarried, fecund remarried women who were married to men in first marriages had significantly more births. Women and men in 1910 faced strong cultural and familial pressures to reproduce, especially if married for the first time. Thus, as we expected, this remarital configuration yielded more children. Perhaps women and men in this configuration were motivated to cement the marriage through childbearing. Selectivity in entering marriage to support future childbearing may also have occurred.
However, the ability to cement the marriage is dependent on the capacity to bear children. Among remarried women, births at an earlier point in the life course demonstrates this capacity, as indicated in the model by the presence of biological children from a prior marriage in the household. Compared with remarried women who had 1–2 children from their prior marriage in the household, the log odds of being infecund were higher among women who had no children from their prior marriage in the household and lower among women who had 3 or more children from their prior union in the household. Among fecund women and compared with the same reference group, women with no children from a prior union had significantly fewer remarital births, whereas women with 3 or more children from a prior union had significantly more. Additionally, women with at least one of their husband’s children from a prior marriage in the household were neither more nor less likely to be in the infecund group, but they had significantly more remarital births than women without at least one of their husband’s children from a prior marriage in the household. Although men with children from a prior marriage had demonstrated a capacity to father children, which might contribute in some way to the observed association, we again caution that selection of remarital partners with preferences for larger or smaller families may have been occurring.12
Socioeconomic factors also influenced fecundity and fertility. Although literacy was not associated with the log odds of being in the infecund group, fecund women who were literate had fewer children than fecund women who were illiterate. Additionally, among fecund women, higher occupational prestige was associated with fewer remarital births, although this association was only marginally significant (p < .10). The findings relating to literacy and occupational prestige are consistent with the multiple-causes framework: higher-status women were more likely to know about and adopt modern methods of contraception and abortion, or to have smaller family size preferences associated with greater investments in children. We also find strong associations with tenant farming. The log odds of being infecund are lower among women residing in tenant versus nontenant farm households. Also, fecund women residing in tenant farm households have significantly more remarital births. These findings are consistent with Tolnay’s (1984, 1999) arguments about the positive pressures of tenancy status on fertility, especially among those on the lower rungs of the agricultural ladder. However, because our data are cross-sectional, findings may also reflect selective migration; infecund or subfecund women may have had to leave tenant farms for other living arrangements and/or occupations.
Geographical factors also shaped fecundability and remarital fertility in expected ways. The log odds of being in the infecund group were higher among urban than other women. Fecund women who lived in towns or cities had significantly fewer remarital births than those who lived in rural areas. Both patterns may reflect voluntary control; nonvoluntary factors associated with the poorer sanitary and disease environments that prevailed in cities and (mill) towns in this period (Elman et al. 2015); or, as noted earlier in the discussion of the findings for tenancy, selective migration. Additionally, compared with women who lived in the South, the log odds of being in the infecund group were significantly higher among women who lived in the West. However, fecund women living in the Midwest or West had significantly fewer remarital births than women living in the South. This latter finding is consistent with evidence that the system of agricultural production prevailing in the South encouraged larger family size (Elman et al. 2015; Tolnay 1984, 1999).
The results indicate that it is important to account for children missing due to death and fostering/aging out: a higher number of missing children due to death and fostering/aging out was associated with reductions in the log odds of being in the infecund group. This finding makes sense given that women had to have been fecund at some point to have children who could be missing. Additionally, among fecund women, a higher number of missing children was associated with a reduction in the number of remarital births (although the association is only marginally significant for missing due to death). This finding suggests that some of the missing children were from the current remarital union.
We conducted three supplemental analyses to evaluate the robustness of our main findings. Key findings are reported in Table 4.
First, we reestimated Model 2 from Table 3 on the subsample residing in the South, where the majority of African Americans lived in 1910 (Table 4, panel A). Our conclusions regarding race-related differences in being in the infecund group are the same as our main findings; the log odds of being in the infecund group are significantly higher among African American than among white women. However, in the South, fecund African American women had the same number of remarital births as fecund white women. These findings are similar to those reported by Tolnay (1981, 1983, 1984) and are consistent with the interpretation that prevailing systems of rural production in the South exerted positive pressure on both white and African American women, while countervailing environmental and disease conditions impaired African American women’s fecundity by involuntary means.
We conducted a second supplemental analysis to evaluate the sensitivity of our findings and conclusions to unobserved children in the household. Overall, 56.80 % of remarital households had all the children the woman had ever borne enumerated in the household at the time of the census. There was little difference in this proportion by race: 56.45 % of white households and 58.12 % of African American households were not missing any children. In Table 4, panel B, Model 1, we present the results from the ZINB model estimated on the subsample of women who had all their children ever born enumerated in their households at the time of the census. In Table 4, panel B, Model 2, we present the results from the ZINB model estimated on the subsample of women who were missing at least one child. We control for the number of missing children due to death and fostering/aging out in Model 2; we do not do so in Model 1 because the number of missing children, by definition, is zero.
Focusing on the subsample of women with no missing children, the pattern of results is identical to our main findings; the log odds of being in the infecund group are significantly higher among African American than white women, and fecund African American women have significantly fewer remarital births than do fecund white women. The pattern for women with missing children resembles that in the southern sensitivity analysis (see Table 4, panel A). Specifically, adjusting for the number of missing children, the log odds of being infecund were higher among African American than white women, but after the fecundity threshold was passed, there was no race difference in the number of remarital births.
The race-related results of the third supplemental analysis are presented in Table 4, panel C. Here, we added a third remarital configuration to the sample—first-married women married to previously married men—which allows us to examine race differences in the population of all remarriages rather than the population of remarried women. The pattern is substantively the same as that in our primary analyses. Results indicate that the log odds of being in the infecund group were significantly higher among African American women relative to white women; among the fecund, African American women had fewer remarital births than white women.
Past studies found strong evidence that multiple causes—voluntary and involuntary influences on fertility linked to economic development, changing norms, and maternal health status—contributed to race-related differences in fertility in the early twentieth century. However, past research focused on samples of younger or first- or ever-married women and did not differentiate remarital fertility patterns. In the early twentieth century, women’s remarriage and fertility patterns reflected the life challenges that they faced, rooted in the biosocial conditions that produced racial disparities in health. In this era, remarriage risk (largely due to the force of mortality and widowhood) and fertility/infertility risk emerged in young adulthood and extended to midlife. Although potential risk in each domain overlapped in virtually all women’s lives in 1910, they most often jointly manifested in the lives of African American women.
We used 1910 IPUMS data to examine fecundity and fertility among the remarried, adjusting for a broad range of individual, marital, socioeconomic, and geographic control variables. Importantly, we also controlled for children missing from the household due to death or fostering/aging out in the ZINB models that we estimated. Based on previous multiple-causes studies, we expected to find that, ceteris paribus, African American women would more likely be infecund than white women and that fecund African American women would also have fewer remarital births. Results supported these expectations, even as other factors influenced fecundity and remarital fertility in expected ways. Supplemental sensitivity analyses both reinforced and nuanced our main findings. For example, in every analysis that we conducted, we found that African American women were more likely than white women to be infecund. However, when we limited our analyses to the South, where the majority of African Americans lived in 1910 and where political-economic conditions supported sharecropping and its strong economic incentives for high fertility (Elman et al. 2015; Tolnay 1984, 1999), we found no race difference in the number of remarital births among fecund women.
Our findings support a multiple-causes interpretation. Consistent with demographic perspectives stressing voluntary control in historical fertility decline, literacy and town and city residence (all associated with increased knowledge about and access to contraception) were associated with fewer remarital births among fecund women. Fecund women with more children from prior unions in their households had more remarital births due to selection/preference or economic need for larger families. Remarried fecund women who wed first-married (as opposed to remarried) men also had more births, perhaps due to selectivity, social expectations, and/or the desire to cement remarriages through childbearing. However, involuntary forces played a role in remarital fertility as well. Infecundity involves sterility, (serial) fetal miscarriage(s), and other physiological factors; subfecundity involves a diminished reproductive capacity due to aging, disease, or stress (McFalls 1900). Women who remarried at older ages were more likely to be in the infecund group and to have fewer remarital births. Adjusting for age at remarriage, duration of remarriage, and remarital configuration, women who were childless in a first marriage—who should have been as or more likely than other women to have a first remarital birth—remained childless as they were significantly less likely to have even one remarital birth. These findings are suggestive of long-term infecundability among a subset of women, regardless of race.
However, although farm tenancy was associated with more remarital births among fecund women, it was also associated with a higher log odds of being infecund, which may have resulted from greater exposure to economic precariousness and health risks, particularly in the South (Elman et al. 2015; Tolnay 1999). Moreover, African American maternal child health around 1910 was inversely related to local tenant farm density (Elman et al. 2015; Humphreys 2001; Tolnay 1984). In the historical era that we examine, most married women in the United States did not prefer childlessness (Morgan 1991), and contraception and abortion were most likely used to stop childbearing at higher parities than to stop it all together or delay it early in marriage. Thus, it is unlikely that our most consistent finding—that African American remarried women had a reduced likelihood of having any remarital birth relative to white women—can be explained by voluntary fertility limitation. Strong Coale-Trussell method evidence of minimal voluntary fertility control among rural African American women around 1910 (Tolnay 1981, 1987)—in an era when most African American women were rural—enhances our conjecture that involuntary factors related to health mattered.
Past studies of race-related fertility differences among younger or first-married women in the early twentieth century, which may underestimate the age-related prevalence of poorer health or “weathering” effects on fertility, concluded that involuntary factors played a significant role around 1910 (Cutright and Shorter 1979; Farley 1966, 1970a, b; McFalls and McFalls 1984; Tolnay 1983, 1987, 1989). These studies proposed that higher African American disease burden, poorer nutrition exacerbated by sharecropping food cycles, and limited access to health care disproportionately increased African American women’s vulnerability and perinatal death. Our findings are in this vein (see also Elman et al. 2015). We believe that additional research is needed to pick up where our study and others have left off and to better empirically specify the differential health-compromising circumstances that impaired remarried, as well as first-married, African American (and white) women’s fecundity and fertility in the early twentieth century.
The authors contributed equally in the preparation of this manuscript. We thank the reviewers for their helpful comments.
Race is partially a proxy for income and captures otherwise unmeasured differential access to housing, nutrition, medical care, and safe environments (Costa 2004).
McFalls (1990) noted that fertility is constrained by intentional control via contraception, abortion, and a low propensity to mate, and by fecundity (the ability to reproduce). Infecundity (total lack of reproductive ability) is due to sterility, fetal miscarriage/pregnancy loss, or the inability to have coitus for physiological reasons; subfecundity is a diminished capacity to reproduce due to aging, disease, or stress.
The African American crude birth rate fell from approximately 50 per 1,000 in 1880 to 35 per 1,000 by 1920 (Farley 1966). Rural fertility decline, commencing around 1910, was the strongest component of the overall African American fertility decline (Farley 1970a; Tolnay 1987).
Higher African American than white mortality rates, in conjunction with higher African American marriage rates (Kiser 1958) at earlier mean ages (Koball 1998; Landale and Tolnay 1991), produced significantly higher rates of widow(er)hood and remarriage than among whites of comparable age (Glick 1949; London and Elman 2001; Preston et al. 1992).
U.S. censuses prior to 1910 did not report marriage order. For example, Tolnay (1981, 1983, 1984) fashioned a series of 1900 PUMS studies to capture the fertility patterns of first-married women as much as possible.
African American life expectancy at birth in 1880 was 30 to 35.5 years, and it remained this low in 1910 (Eblen 1974; Farley 1965; Ruggles 1994). Approximately 90 % of African Americans lived in the South in 1880; the proportion declined to 85 % by 1920 (Engerman 1977).
The South’s tenant farm labor system promoted early marriage and childbearing among those without the means to buy land in young adulthood (Bloome and Muller 2015; Landale 1989; Landale and Tolnay 1991). It was uncommon for African American sharecroppers to move into ownership in the early twentieth century (Alston and Ferrie 2005).
In the Coale-Trussell model, the parameter m represents the degree to which observed age-specific fertility rates differ from the age-specific rates observed in a standard natural fertility population; m values under .3 indicate little, if any, voluntary fertility control at higher parities. Tolnay (1987:213) found a “modest” role for voluntary fertility limitation among nonfarm African Americans in the period 1905–1910, but “only very weak evidence of fertility control among rural blacks as . . . the m value of .338 for farm women is very close to the range typically associated with natural fertility populations.” There was a hint of parity-specific control among African American rural women in 1880, but this had not increased by 1900 (Tolnay 1981).
In the 1910 IPUMS, there was considerable African American overreporting of widowhood versus singlehood (Preston et al. 1992) and underreporting of remarital versus first-married status (Kramarow 1995).
Among native whites in 1910 widowed by age 40–44, 50 % of men and 40 % of women remarried (Kramarow 1995). Among native whites divorced by age 40–44 in 1910, 84 % of men and 82 % of women remarried (Kramarow 1995).
The 1910 census was the first to ask about remarriage. Enumerators recorded “M1” for persons in first marriages and “M2” for persons in second or subsequent marriages (U.S. Bureau of the Census 1910:29). Approximately 6 % were given a number greater than 2 (i.e., “M3” or “M4”); another 4 % of married respondents were recorded as “M” without a number attached (Kramarow 1995). Kramarow (1995) estimated that remarriages were underreported by no more than 8.3 %.
Children from prior marriages in households increase the number of remarital births, which may reflect shared but unmeasured cultural influences on the range of proximate determinants that shape fertility patterns (Bongaarts and Potter 1983).